The area of the relative operating characteristic (ROC) curve is a criterion-free parameter-free distributionindependent index of the diagnostic performance of a test. The area of the ROC curve is equivalent to the probability of making the correct choice in a two-alternative forced-choice decision task. In this paper, we describe several methods for computing the area of the ROC curve from data distributions, we provide algorithms for computing confidence intervals on ROC curve areas and for performing statistical comparisons of ROC curve areas, and we show how the ROC curve areas can be used to evaluate the efficacy of different diagnostic tests or data reduction procedures. As an example of the application of the different techniques for computing the ROC curve area, we compare three distinctly different data reduction procedures applied to static visual field data from glaucoma patients, glaucoma suspects, and normal controls. Based on ROC curve areas, it is concluded that none of the three procedures is significantly superior to the others at discriminating glaucoma patients or glaucoma suspects from normals.
ASJC Scopus subject areas
- Atomic and Molecular Physics, and Optics
- Engineering (miscellaneous)
- Electrical and Electronic Engineering